CtrlK
Aleph Alpha is a German artificial intelligence company headquartered in Heidelberg, Germany. Founded in 2019 by Jonas Andrulis and Samuel Weinbach, the company develops sovereign AI solutions for enterprises and government institutions, with a strong emphasis on European data sovereignty, regulatory compliance, and explainability. Aleph Alpha gained international attention as one of Europe's most prominent AI startups, raising over $500 million in a high-profile Series B round in November 2023. The company initially built its own family of large language models called Luminous before pivoting in 2024 to become an enterprise AI platform provider through its PhariaAI operating system.
Aleph Alpha GmbH was established in January 2019 in Heidelberg, Germany, by Jonas Andrulis and Samuel Weinbach. Andrulis, the company's CEO, had previously spent three years as a senior AI R&D manager at Apple's Special Projects Group (SPG). Before Apple, he founded Pallas Ludens, a machine learning and computer vision company, and held consulting roles at Deloitte and msgGillardon AG. He studied industrial engineering at the Karlsruhe Institute of Technology (KIT), where his thesis focused on Bayesian networks. Weinbach, who became co-chief research officer, brought expertise from Deloitte in AI consulting and implementation.
The founding mission was to research and develop sovereign, human-centric AI technologies that would give European organizations an alternative to American AI providers. From the outset, Aleph Alpha positioned itself as a company building AI for businesses and government agencies rather than consumer-facing products.
In November 2020, the company secured its seed round of approximately 5.3 million euros from LEA Partners, 468 Capital, and Cavalry Ventures.
In July 2021, Aleph Alpha raised approximately 23 million euros (around $27 million) in a Series A round co-led by Earlybird VC, Lakestar, and UVC Partners. The funding brought total investment to roughly 28.3 million euros and allowed the company to scale its research into large language models.
During this period, Aleph Alpha developed and launched the Luminous model family, a series of multilingual transformer-based LLMs offered in three sizes. The company also published MAGMA, a multimodal model architecture that could process both text and images, which was integrated into the Luminous models to give them vision-language capabilities.
In 2022, Aleph Alpha opened its alpha ONE data center at the GovTech Campus in Berlin. This facility would later become the first site in Europe to deploy Cerebras CS-3 AI supercomputers and was marketed as one of the fastest commercial AI data centers in Europe.
On November 6, 2023, Aleph Alpha announced a funding round of more than $500 million (approximately 470 million euros at the exchange rate at the time of signing). The round drew significant media attention and was widely described as one of the largest AI funding rounds in European history.
However, the structure of the round was more nuanced than the headline figure suggested. A detailed breakdown revealed three components:
The round was led by the Innovation Park Artificial Intelligence (Ipai), Bosch Ventures, and the Schwarz Group (the parent company of retail chains Lidl and Kaufland, and Europe's largest retailer by revenue). Additional new investors included Hewlett Packard Enterprise, SAP, Christ&Company Consulting, and Burda Principal Investments. Existing investors from previous rounds also participated.
The Schwarz Group's involvement proved strategically pivotal. Through its digital subsidiary Schwarz Digits, the retail conglomerate operates STACKIT, a sovereign European cloud infrastructure platform. This investment marked the beginning of a deep integration between Aleph Alpha's AI technology and STACKIT's cloud infrastructure.
By mid-2024, Aleph Alpha shifted its strategy away from competing directly as a foundation model provider against American labs such as OpenAI, Anthropic, and Google DeepMind. The company acknowledged that the economics of training frontier models from scratch were extremely challenging for a European startup, particularly given the massive capital expenditures required for compute.
Instead, Aleph Alpha repositioned itself as a provider of enterprise AI infrastructure, launching PhariaAI as a full-stack "operating system for generative AI." This platform was designed to allow enterprises and government agencies to deploy, govern, and operate AI systems while maintaining data sovereignty and regulatory compliance.
In March 2024, Aleph Alpha acquired the software assets of Berlin-based Lengoo, an AI-driven enterprise translation platform. As part of the acquisition, Lengoo's co-founder and CEO Christopher Kranzler was appointed Vice President of Product at Aleph Alpha, along with over 20 engineers, product managers, and designers from the Lengoo team.
The company also released Pharia-1-LLM, a 7-billion parameter open-weight model family, in 2024, signaling a shift from proprietary model development toward contributing to the open-source AI ecosystem while focusing commercial efforts on the platform layer.
In the second half of 2025, Aleph Alpha underwent significant leadership and organizational changes. Reto Sporri, the former CEO of Lidl's commerce division, joined as co-CEO alongside Andrulis in mid-2025. In October 2025, Andrulis announced he would step down as CEO.
Effective January 1, 2026, Andrulis transitioned to the role of Chairman of the Advisory Board. Ilhan Scheer, who had been serving as Chief Growth Officer, was appointed co-CEO alongside Sporri to lead the company going forward. Carsten Dirks, who had been responsible for operational business, also departed the company.
Alongside the leadership change, Aleph Alpha carried out a reduction of approximately 50 employees, corresponding to roughly 17 percent of the workforce. The cuts primarily affected engineering and research teams, particularly divisions focused on improving model performance, reflecting the company's pivot away from in-house model development toward platform and integration work.
In April 2025, Aleph Alpha completed its second acquisition in less than a year, purchasing thingsTHINKING, a Karlsruhe-based semantic AI company. The acquisition brought thingsTHINKING's product semantha, a platform for semantic text analysis used in contract review and requirements engineering, into Aleph Alpha's enterprise portfolio.
The Schwarz Group continued to expand its influence over the company, reportedly seeking to take over the shareholding previously held by Bosch Ventures, deepening the alignment between Aleph Alpha's product roadmap and the Schwarz Digits ecosystem.
The Luminous family was Aleph Alpha's flagship series of large language models, offered through the company's API from 2022 onward. The models followed a decoder-only autoregressive transformer architecture and used rotary positional embeddings (RoPE). They were trained on a curated multilingual corpus covering English, German, French, Italian, and Spanish.
| Model | Parameters | Context Length | Training Tokens | Multimodal | Key Characteristics |
|---|---|---|---|---|---|
| Luminous-Base | 13 billion | 2,048 | ~400 billion | Yes | Fastest and most cost-effective; suitable for straightforward tasks |
| Luminous-Extended | 30 billion | 2,048 | ~450 billion | Yes | Mid-range option balancing capability and cost |
| Luminous-Supreme | 70 billion | 2,048 | ~588 billion | Yes | Largest and most capable; best for complex reasoning |
All three base models were also available in instruction-finetuned variants called Luminous-control models, which were optimized for following user instructions in conversational and task-oriented settings.
A distinguishing feature of the Luminous family was built-in multimodal capability. Drawing on Aleph Alpha's earlier research on MAGMA (Multimodal Augmentation of Generative Models through Adapter-based finetuning), the Luminous models could process and reason over both text and images. Users could ask questions about images in natural language and receive AI-generated answers, going beyond simple object recognition.
Another key differentiator was the integration of AtMan (Attention Manipulation), an explainability method that produces relevance maps showing which input tokens influenced the model's output. AtMan works by manipulating attention mechanisms across all transformer layers and measuring the resulting differences in output probabilities. This allowed Aleph Alpha to offer a degree of transparency unusual among LLM providers, which was particularly attractive to government and regulated-industry customers.
In 2024, Aleph Alpha released Pharia-1-LLM, a smaller, more efficient model family published under the Open Aleph License for non-commercial research and educational use. The family included two 7-billion parameter variants:
| Model Variant | Parameters | License | Description |
|---|---|---|---|
| Pharia-1-LLM-7B-control | 7 billion | Open Aleph License | Instruction-tuned model with length-controlled responses; optimized for German, French, and Spanish |
| Pharia-1-LLM-7B-control-aligned | 7 billion | Open Aleph License | Further refined through KTO (Kahneman-Tversky Optimization) using human and LLM preferences |
Both models used RoPE positional embeddings and were trained on carefully curated data in compliance with EU regulations, including copyright and data privacy laws. The models were designed for domain-specific applications, with particular strength in automotive and engineering use cases.
In early 2025, Aleph Alpha introduced a novel tokenizer-free (T-Free) model architecture called the Hierarchical Autoregressive Transformer (HAT). Traditional language models rely on tokenizers that split text into subword units, but this approach introduces efficiency penalties for languages with different morphological structures than English.
The T-Free architecture operates directly on raw UTF-8 bytes, using a vocabulary of only 256 entries. Text is first split into words using simple rules; a small encoder module then processes the bytes within each word and maps them to a word embedding. The resulting sequence of word embeddings is then processed by a larger backbone model.
Aleph Alpha reported that this approach could reduce training costs and carbon footprint by up to 70 percent for morphologically rich languages such as Finnish, compared to token-based alternatives. In benchmarks, the HAT architecture achieved approximately 40 percent better compression in German and 16 percent better compression in English compared to Llama 3.1 8B.
Two open models were released based on this architecture: TFree-HAT-Pretrained-7B-Base (pre-trained in English and German) and Llama-TFree-HAT-Pretrained-7B-DPO (a post-trained variant).
PhariaAI is Aleph Alpha's enterprise-grade operating system for generative AI, introduced in late 2024 as the centerpiece of the company's strategic pivot. It unifies deployment, governance, explainability, and compliance into a single stack for production teams.
Key capabilities of PhariaAI include:
PhariaAI is available as PhariaAI-as-a-Service through the STACKIT cloud, providing regulated enterprises and public-sector customers with a fully European AI stack.
Aleph Alpha's total disclosed funding exceeds $530 million across four rounds.
| Round | Date | Amount | Key Investors |
|---|---|---|---|
| Seed | November 2020 | ~5.3 million euros | LEA Partners, 468 Capital, Cavalry Ventures |
| Series A | July 2021 | ~23 million euros ($27 million) | Earlybird VC, Lakestar, UVC Partners |
| Series B | November 2023 | ~470 million euros ($500+ million) | Schwarz Group, Bosch Ventures, Ipai, HPE, SAP, Burda Principal Investments, Christ&Company |
| Total | ~$530+ million |
The Series B round's structure included 110 million euros in equity financing, 300 million euros in research funding for Aleph Alpha Research, and 60 million euros in order commitments. Journalists covering the round noted that the equity component alone valued the company at an estimated $500 to $625 million.
Aleph Alpha has established itself as a key AI supplier to the German public sector. Notable government relationships include:
In May 2024, Aleph Alpha and Cerebras Systems announced a multi-year partnership to train generative AI models for the German Armed Forces. The collaboration involved deploying Cerebras CS-3 AI supercomputers at Aleph Alpha's alpha ONE data center at the GovTech Campus in Berlin.
Aleph Alpha's enterprise customers span regulated industries including automotive, engineering, financial services, and healthcare. The company's emphasis on data sovereignty and explainability makes it attractive to organizations in sectors where regulatory compliance is a primary concern.
Through its Pharia Industrial Suite, which incorporates technology from the thingsTHINKING and Lengoo acquisitions, Aleph Alpha offers specialized solutions for contract analysis, requirements engineering, and multilingual document processing.
The most significant partnership is with Schwarz Digits and its STACKIT cloud platform. PhariaAI runs natively on STACKIT, creating an end-to-end sovereign AI stack that is entirely hosted within German and European data centers. This partnership positions Aleph Alpha as the AI layer within a broader German alternative to U.S. hyperscalers such as Amazon Web Services and Microsoft Azure.
Additional technology partnerships include collaborations with AMD for hardware optimization and with Cerebras Systems for high-performance AI training infrastructure.
Aleph Alpha has made European data sovereignty a core pillar of its business strategy, differentiating itself from American AI providers that typically process data on U.S.-based infrastructure.
All Aleph Alpha systems are hosted within Germany, ensuring that data remains under European jurisdiction at all times. This is a critical requirement for public-sector and enterprise customers bound by GDPR restrictions on cross-border data transfers, particularly following the Schrems II ruling that invalidated the EU-U.S. Privacy Shield framework.
Aleph Alpha has positioned itself as an early adopter of the EU AI Act, the world's first comprehensive regulatory framework for artificial intelligence. The company asserts that its models are designed to meet the Act's requirements from the outset, including obligations around transparency, documentation, and risk management for general-purpose AI models.
Aleph Alpha was among the signatories of the EU's General-Purpose AI Code of Practice, joining companies including Cohere, IBM, Mistral AI, and OpenAI in committing to a voluntary framework for responsible development of general-purpose AI systems.
The concept of "sovereign AI" is central to Aleph Alpha's market positioning. In this context, sovereignty refers to the ability of organizations and governments to maintain full control over their AI infrastructure, data, and decision-making processes without dependency on foreign technology providers. This resonates particularly with European defense agencies, public-sector organizations, and companies in critical infrastructure sectors.
Aleph Alpha operates in a competitive landscape that includes both American frontier AI labs and European AI companies.
The primary competitive challenge comes from U.S.-based AI labs including OpenAI, Anthropic, Google DeepMind, and Meta AI. These organizations have access to significantly larger capital pools and compute resources, allowing them to train larger frontier models. Aleph Alpha's pivot away from direct model competition in 2024 was in part a recognition that matching the scale of American labs was not economically viable for a European startup.
Within Europe, the most prominent competitor is Mistral AI, a French AI company that has pursued a different strategy by releasing high-performance open-weight models. Mistral AI raised 1.7 billion euros in September 2025, reaching a valuation of approximately 11.7 billion euros, significantly outpacing Aleph Alpha in both funding and valuation.
Where Mistral focuses on building competitive foundation models, Aleph Alpha has differentiated by emphasizing enterprise integration, explainability, and sovereign deployment. Other European AI efforts include the OpenEuroLLM project and various national AI initiatives.
Aleph Alpha's competitive advantage lies not in model performance benchmarks but in its integrated platform approach, regulatory compliance features, and deep partnerships with German government agencies and the Schwarz Group's cloud infrastructure. For customers where data sovereignty, auditability, and EU compliance are primary requirements, this positioning fills a market gap that American providers cannot easily address.
| Name | Role | Period | Background |
|---|---|---|---|
| Jonas Andrulis | Founder, CEO; later Chairman of Advisory Board | 2019 to 2025 (CEO); 2026 onward (Chairman) | Former Apple AI R&D Manager; KIT graduate; founder of Pallas Ludens |
| Samuel Weinbach | Co-Founder, Co-Chief Research Officer | 2019 onward | Former Deloitte AI consultant; leads Aleph Alpha Research |
| Reto Sporri | Co-CEO | Mid-2025 onward | Former CEO of Lidl's commerce division |
| Ilhan Scheer | Co-CEO | January 2026 onward | Previously Chief Growth Officer at Aleph Alpha |
| Detail | Information |
|---|---|
| Legal Name | Aleph Alpha GmbH |
| Founded | January 2019 |
| Headquarters | Heidelberg, Germany |
| Founders | Jonas Andrulis, Samuel Weinbach |
| Employees | Approximately 350 (as of early 2026) |
| Total Funding | $530+ million |
| Key Products | PhariaAI, Luminous model family, Pharia-1-LLM, T-Free architecture |
| Key Markets | Government, defense, automotive, engineering, financial services |
| Research Subsidiary | Aleph Alpha Research |
| Year | Event |
|---|---|
| January 2019 | Aleph Alpha GmbH founded in Heidelberg by Jonas Andrulis and Samuel Weinbach |
| November 2020 | Seed round of 5.3 million euros from LEA Partners, 468 Capital, and Cavalry Ventures |
| July 2021 | Series A round of 23 million euros led by Earlybird VC, Lakestar, and UVC Partners |
| 2022 | Alpha ONE data center opened at the GovTech Campus in Berlin |
| 2022 | Luminous model family launched publicly |
| January 2023 | AtMan explainability method published (arXiv paper) |
| November 2023 | Series B funding of 470 million euros ($500+ million) announced |
| March 2024 | Acquisition of Lengoo's software assets |
| May 2024 | Partnership with Cerebras Systems announced; CS-3 supercomputers deployed at alpha ONE |
| Mid-2024 | Strategic pivot from foundation model provider to enterprise AI platform |
| 2024 | Pharia-1-LLM 7B open-weight models released |
| Late 2024 | PhariaAI enterprise platform launched |
| Early 2025 | T-Free (Hierarchical Autoregressive Transformer) architecture introduced |
| April 2025 | Acquisition of thingsTHINKING (semantic AI startup) |
| Mid-2025 | Reto Sporri joins as co-CEO alongside Andrulis |
| October 2025 | Jonas Andrulis announces departure as CEO |
| Late 2025 | Workforce reduction of approximately 50 employees (~17% of staff) |
| January 2026 | Reto Sporri and Ilhan Scheer become co-CEOs; Andrulis transitions to Chairman of Advisory Board |